Time series forecasting is the use of a model to predict the future values of a time series from its history. Methods range from simple rules, such as assuming the next value equals the last one, through exponential smoothing and ARIMA, to neural networks such as long short-term memory models. A forecast is normally given with a range of uncertainty that widens as the horizon lengthens. It is the predictive branch of time series analysis.
Forecasts are judged on data not used to build the model. Because order matters, the test period must come after the training period, as in walk-forward optimization. In markets, the direction of returns is very hard to predict, and a naive forecast based on the random walk is difficult to beat. Volatility, trading volume and many economic indicators are more predictable. A small forecasting edge can still be valuable, but only if it survives trading costs.